Joint Long-Term Base Station Activation and Short-Term Beamforming: From Optimization to Learning
报告专家：林静然
报告时间：2022年9月22日星期四晚上8：30-9：30
报告地点：腾讯会议：898-495-592
摘要：Base station (BS) activation is a widely-used approach to alleviate the system power cost for device maintenance. However, frequently switching on/off BSs may also introduce extra power and signaling costs. This motivates us to limit the BS switching frequency when performing BS activation. To address this, we study a joint long-term BS activation and short-term beamforming problem in a network where multiple multi-antenna BSs cooperatively serve multiple single-antenna users. By jointly optimizing the active BSs and the associated transmit beamformers in different time slices, we aim to well balance the transmit power, the maintenance power and the switching power. Three algorithms, including two optimization (OP)-based algorithms and one reinforcement learning~(RL)-based algorithm, are designed for the problem. One OP-based algorithm solves the problem in an offline manner, which collects all channel state information (CSI), and solves the active BSs and the beamformers in different time slices in one shot; another OP-based algorithm is online, which determines the active BSs and the beamformers time slice by time slice, utilizing current CSI and previous active BSs. After fitting the problem into the actor-critic~(AC) framework, the RL-based algorithm learns an efficient long-term BS activation policy. Finally, extensive numerical simulations demonstrate the efficacies of the algorithms.
专家简介：林静然，1978年生，重庆人。1997年9月至2007年6月，就读于电子科技大学通信与信息工程学院，相继获“计算机通信”专业工学学士学位和“信号与信息处理”专业工学博士学位。毕业后留校工作至今，2007年7月被聘为讲师，2009年7月被聘为副教授，2020年7月遴选为博士生导师，2021年5月被聘为教授。2012年1月至2013年1月为美国明尼苏达大学（双城校区）罗智泉教授[Prof. Zhi-Quan (Tom) Luo]研究团队的访问学者。近五年研究主要集中于电磁认知与应用，大规模无线资源优化等领域，主持各级项目30余项，共发表学术论文60余篇，申报专利70余项（授权50余项），出版著作4部。论文主要发表在IEEE TIFS、IEEE TSP、IEEE TWC, IEEE TVT、Digital Signal Processing (DSP)、IEEE Signal Processing Letters (SPL)等SCI期刊，以及ICASSP, GLOBECOM, EUSIPCO等IEEE通信信号处理会议，单篇最高引用141次（Google Scholar，截止2022年9月）。
